Accessing bigger datasets in R using SQLite and dplyr

Transcription

1 Accessing bigger datasets in R using SQLite and dplyr Amherst College, Amherst, MA, USA March 24, 2015 [email protected]

2 Thanks to Revolution Analytics for their financial support to the Five College Statistics Program and the Biostatistics Program at UMass/Amherst to Nick Reich, Emily Ramos, Kostis Gourgoulias, and Andrew Bray (co-organizers of the Meetup) suggestions of meeting topics, speakers and locations always welcomed

3 Upcoming meetups working with geographic data in R 3/26 at 7:00pm DataFest 3/27-3/29 learn about plotly 4/2 at 4:00pm updates from RStudio: J.J. Allaire 7/15 (save the date!) statistical analysis of big genetics and genomics data Xihong Lin, 11/16 (save the date!) using docker for fun and profit anyone willing?

4 Acknowledgements joint work with Ben Baumer (Smith College) and Hadley Wickham (Rice/RStudio) supported by NSF grant (building a community around modeling, statistics, computation and calculus) more information at

5 Prelude New Frontier in statistical thinking (Chamandy, JSM 2013) teaching precursors to big data/foundations of data science as an intellectually coherent theme (Pruim, Calvin College) growing importance of computing and ability to think with data (Lambert, Google) key capacities in statistical computing (Nolan and Temple Lang, TAS 2010) Statistics and the modern student (Gould, ISR 2010) Building precursors for data science in the first and second courses in statistics (Horton, Baumer, and Wickham, 2015)

6 Prelude (cont.) How to accomplish this? start in the first course build on capacities in the second course develop more opportunities for students to apply their knowledge in practice (internships, collaborative research, teaching assistants) new courses focused on Data Science today s goal: helping DataFesters grapple with larger datasets

7 Data Expo 2009 Ask students: have you ever been stuck in an airport because your flight was delayed or cancelled and wondered if you could have predicted it if you d had more data? (Wickham, JCGS, 2011)

8 Data Expo 2009 dataset of flight arrival and departure details for all commercial flights within the USA, from October 1987 to April 2008 (but we now have through the begininning of 2015) large dataset: more than 180 million records aim: provide a graphical summary of important features of the data set winners presented at the JSM in 2009; details at

9 Airline Delays Codebook (abridged) Year Month DayOfMonth DayOfWeek 1=Monday DepTime departure time UniqueCarrier OH = Comair, DL = Delta, etc. TailNum plane tail number ArrDelay arrival delay, in minutes Origin BDL, BOS, SFO, etc. Dest Full details at

10 Data Expo 2009 winners

11 Data Expo 2009 winners

12 Data Expo 2009 winners

13 Data Expo 2009 winners

14 Data Expo 2009 winners

15 Goal facilitate use of database technology for instructors and students without training in this area demonstrate how to use SQLite in conjunction with dplyr in R to achieve this goal help enable the next generation of data scientists

16 Background on databases and SQL relational databases (invented in 1970) like electronic filing cabinets to organize masses of data (terabytes) fast and efficient useful reference: Learning MySQL, O Reilly 2007 Horton, Baumer, and Wickham (2015, in press),

17 Client and server model server: manages data client: ask server to do things use R as both the client and the server

18 SQL Structured Query Language special purpose programming language for managing data developed in early 1970 s standardized (multiple times) most common operation is query (using SELECT)

19 Why SQLite? advantage: free, quick, dirty, simple (runs locally), database can be copied from one system (Mac OS X) to Windows or Linux disadvantage: not as robust, fast, or flexible than other free alternatives such as MySQL (which run remotely) For personal use, SQLite is ideal (and is what I will demonstrate today). For a class, I d recommend MySQL.

20 Creating the airline delays database 1 download the data (30gb uncompressed) 2 load the data 3 add indices (to speed up access to the data, takes some time) 4 establish a connection (using src sqlite()) 5 start to make selections (which will be returned as R objects) using dplyr package 6 features lazy evaluation (data only accessed when needed)

21 Key idioms for dealing with big(ger) data select: subset variables filter: subset rows mutate: add new columns summarise: reduce to a single row arrange: re-order the rows Hadley Wickham, bit.ly/bigrdata4

22 Accessing the database (see test-sqlite.rmd) my_db <- src_sqlite(path="ontime.sqlite3") # link to some useful tables flights <- tbl(my_db, "flights") airports <- tbl(my_db, "airports") airlines <- tbl(my_db, "airlines") airplanes <- tbl(my_db, "airplanes")

23 Accessing the database (see test-sqlite.rmd) > airports Source: sqlite [ontime.sqlite3] From: airports [3,376 x 7] IATA Airport City State 1 00M Thigpen Bay Springs MS 2 00R Livingston Municipal Livingston TX 3 00V Meadow Lake Colorado Springs CO 4 01G Perry-Warsaw Perry NY 5 01J Hilliard Airpark Hilliard FL 6 01M Tishomingo County Belmont MS Variables not shown: Country (chr), Latitude (dbl), Longitude (dbl)

24 Accessing the database (see test-sqlite.rmd) airportcounts <- flights %>% filter(dest %in% c( ALB, BDL, BTV )) %>% group_by(year, DayOfMonth, DayOfWeek, Month, Dest) %>% summarise(count = n()) %>% collect() # this forces the evaluation airportcounts <- airportcounts %>% mutate(date = ymd(paste(year, "-", Month, "-", DayOfMonth, sep=""))) head(airportcounts)

25 GROUP BY ## Source: local data frame [6 x 7] ## Groups: Year, DayOfMonth, DayOfWeek, Month ## ## Year DayOfMonth Day Month Dest count Date ## ALB ## BDL ## BTV ## ALB ## BDL ## BTV

26 SQL commands ## Year Month DayOfMonth Day DepTime CRSDepTime ArrTime CRS ## Variables not shown: UniqueCarrier (chr), FlightNum (int ## ActualElapsedTime (int), CRSElapsedTime (int), AirTime ( ## (int), DepDelay (int), Origin (chr), Dest (chr), Distanc tbl(my_db, sql("select * FROM flights LIMIT 4")) # or tbl(my_db, "flights") %>% head() ## Source: sqlite [ontime.sqlite3] ## From: <derived table> [?? x 29] ## ## ## ## ## ## (int), TaxiOut (int), Cancelled (int), CancellationCode ## (chr), CarrierDelay (chr), WeatherDelay (chr), NASDelay ## SecurityDelay (chr), LateAircraftDelay (chr)

27 Which month is it best to travel (airline averages/bdl)? 30 AvgArrivalDelay

28 Which day is it best to travel (airline averages from BDL)? 30 Avg Delay by Airline (in minutes)

29 Setup on your own machine (data from 2014) Update existing packages: update.packages() Install necessary packages: RSQLite, dplyr, tidyr, mosaic, knitr, nycflights13, lubridate, igraph, markdown Create a new project in RStudio and directory to store the data and datab Download files: see list at Set up the database: run load-sqlite.r Take it for a spin: run test-sqlite.rmd

30 Next steps Want more data? See the additional instructions to access more than just the data from 2014 See also: the dplyr package vignette on databases See also: the precursors paper (in press) at

31 Accessing bigger datasets in R using SQLite and dplyr Amherst College, Amherst, MA, USA March 24, 2015 [email protected]